OpenRouter shipped a version-agnostic alias endpoint — ~z-ai/glm-latest — that always resolves to the newest GLM model from Z.ai.
OpenRouter added ~z-ai/glm-latest, a rolling alias that automatically redirects API calls to the latest released GLM model. Currently it points to z-ai/glm-5.3 (1M context window, reasoning model with configurable effort), but when Z.ai ships GLM 5.4, 5.5, or whatever comes next, the alias moves with it — no code change required.
Pricing stays identical to the pinned version: $0.14/1M input tokens, $0.44/1M output tokens, with cached input at $0.026/1M. The model supports text-only input/output, reasoning with effort levels (low, high, max), tool calls, and structured output.
Why a builder cares: Pinning model versions is the responsible default, but having a -latest alias in your staging/experimental pipeline means you get new model capabilities the day they land — without touching config, CI, or env vars. One endpoint to update when you're ready to cut over; one endpoint to watch if you want to catch regressions before they hit production.